dihedralsoup.blogspot.com

machinedlearnings.com

Machined Learnings: March 2014

http://www.machinedlearnings.com/2014_03_01_archive.html

Cue Butlerian Jihad in 3, 2, 1, . Sunday, March 9, 2014. I spent Christmas break working on some matrix factorization related ideas. There were two things I was wondering about: first, whether dropout is a good regularizer for discriminative low-rank quadratics; and second, how to do the analog of GeV representation learning. Startups have popularized the idea of failing fast. Since most business ideas are also not good ideas. The idea of the minimum viable product. It should work allowed me to power thr...

machinedlearnings.com

Machined Learnings: December 2014

http://www.machinedlearnings.com/2014_12_01_archive.html

Cue Butlerian Jihad in 3, 2, 1, . Monday, December 15, 2014. With a new venue and a deep attitude, NIPS was a blast this year, kudos to the organizers. Let's start with the “talk of the conference”. I mean this in the spirit of Time's “Man of the Year”, i.e., I'm not condoning the content, just noting that it was the most impactful. And of course the winner is . Ilya Sutsveker's talk Sequence to Sequence Learning with Neural Networks. Now for my personal highlights. First let me point out that the co...

machinedlearnings.com

Machined Learnings: September 2014

http://www.machinedlearnings.com/2014_09_01_archive.html

Cue Butlerian Jihad in 3, 2, 1, . Wednesday, September 24, 2014. I have a post on sub-linear debugging on Microsoft's machine learning blog. If you are interested in hitting the Doherty threshold. In Machine Learning, read the whole thing. Posted by Paul Mineiro. Subscribe to: Posts (Atom). If you are looking for full-time employment in applied dialogue systems, drop me a line at. Mbox{pmineiro}@ mbox{microsoft} ldotp mbox{com}$. Mbox{paul} ldotp mbox{mineiro}@ mbox{gmail} ldotp mbox{com}$.

machinedlearnings.com

Machined Learnings: Paper Reviews vs. Code Reviews

http://www.machinedlearnings.com/2015/08/paper-reviews-vs-code-reviews.html

Cue Butlerian Jihad in 3, 2, 1, . Monday, August 10, 2015. Paper Reviews vs. Code Reviews. Because I'm experiencing the NIPS. Submission process right now, the contrast with the ICLR. The traditional model is from an age where papers were actual physical objects that were sent (via snail mail! Incremental review. Rarely a programmer develop a complicated piece of software de novo. Most reviews are about relatively small changes to large pieces of software. To ease the cognitive burden on the ...Modular u...

machinedlearnings.com

Machined Learnings: May 2015

http://www.machinedlearnings.com/2015_05_01_archive.html

Cue Butlerian Jihad in 3, 2, 1, . Monday, May 11, 2015. The selection of accepted papers and invited talks was extremely sensible given the above context: deep learning papers were definitely in the majority, but there were also interesting papers leveraging eigensystems. There were many good papers, so check out the entire schedule. But these caught my eye. Neural Machine Translation by Jointly Learning to Align and Translate. The result in this paper. NICE: Non-linear Independent Components Estimation.

machinedlearnings.com

Machined Learnings: August 2014

http://www.machinedlearnings.com/2014_08_01_archive.html

Cue Butlerian Jihad in 3, 2, 1, . Tuesday, August 26, 2014. More Deep Learning Musings. Yoshua Bengio, one of the luminaries of the deep learning community, gave multiple talks about deep learning at ICML 2014 this year. I like Bengio's focus on the statistical aspects of deep learning. Here are some thoughts I had in response to his presentations. So is it true? Rdquo; but I don't think so. I suspect something extra is happening here, but that's just a guess, and I certainly don't understand it. So for ...

machinedlearnings.com

Machined Learnings: February 2015

http://www.machinedlearnings.com/2015_02_01_archive.html

Cue Butlerian Jihad in 3, 2, 1, . Saturday, February 28, 2015. Wages and the Immigration Debate. It's a reasonable question. According to the BLS. One possibility is that compensation has gone up, but they aren't measuring correctly. That table says “total compensation”, which the footnote says “Includes wages, salaries, and employer costs for employee benefits.” So I suspect (hope! This explanation still leaves open the question: “why don't employers just skip all that stuff, have dumpy offices wi...

machinedlearnings.com

Machined Learnings: January 2015

http://www.machinedlearnings.com/2015_01_01_archive.html

Cue Butlerian Jihad in 3, 2, 1, . Thursday, January 15, 2015. My New Year's resolution is to be less afraid of non-convex optimization. Statistically there is a high likelihood. That I will return to only optimizing convex losses by February :). But here's a fun paper along these lines in the meantime, Deep Fried Convnets. The idea here is to use a fast kernel approximation. I see Deep Fried Convnets along the same lines as John Hershey's Deep Unfolding. Ideas for neural networks, in that one starts with...

machinedlearnings.com

Machined Learnings: May 2014

http://www.machinedlearnings.com/2014_05_01_archive.html

Cue Butlerian Jihad in 3, 2, 1, . Saturday, May 3, 2014. The Most Causal Observer. David K. Park. Recently had a guest post. On Gelman's blog. You should read it. The tl;dr. Is ` Big Data is a Big Deal, but causality is important and not the same as prediction.'. I agree with the basic message: causality is important. For example, why predict the fraud risk of a credit card transaction? Posted by Paul Mineiro. Subscribe to: Posts (Atom). Mbox{pmineiro}@ mbox{microsoft} ldotp mbox{com}$. AI Winter Is Comi...

machinedlearnings.com

Machined Learnings: January 2014

http://www.machinedlearnings.com/2014_01_01_archive.html

Cue Butlerian Jihad in 3, 2, 1, . Sunday, January 12, 2014. It's a safe bet that, for the near future, data will continue to accumulate in clustered file systems such as HDFS, powered by commodity multicore CPUs with ethernet interconnect. Such clusters are relatively inexpensive, fault-tolerant, scalable, and have an army of systems researchers working on them. Is the Coates et. al. result a bespoke solution for deep learning? Interestingly, Canny and Zhao. So here is the current state of affairs as I s...